Modeling the impact of Python and R packages using dependency and contributor networks

This paper develops methods to estimate the factors that affect the impact of open-source software (OSS), measured by number of downloads, with a study of Python and R packages. The OSS community is characterized by a high level of collaboration and sharing which results in interactions between contributors as well as packages due to reuses. We use data collected from Depsy.org about the development activities of Python and R packages, and generate the dependency and contributor networks. We develop three Quasi-Poisson models for each of the Python and R communities using network characteristics, as well as author and package attributes. We find that the more derivative a package is (the more dependencies it has), the less likely it is to have a high impact. We also show that the centrality of a package in the dependency network measured by the out-degree, closeness centrality, and pagerank has a significant effect on its impact. Moreover, the closeness and weighted degree centralities of the developers in the Python and R contributor networks play an important role. We also find that introducing network features to a baseline model using only package features (e.g., number of authors, number of commits) improves the performance of the models.